A machine learning-based intelligent property operation collaborative scheduling method and system

Abstract

The present application relates to the technical field of machine learning, in particular to a kind of wisdom property operation collaborative scheduling method and system based on machine learning, in which, wisdom property operation collaborative scheduling method based on machine learning includes: obtaining the pending work order in current dispatch pool;Emergency of each pending work order is evaluated;The scale of the same kind task in pending work order is determined;Based on the scale of the same kind task and the degree of urgency, the size of attention window is adaptively set;In the attention window, pending work order is prioritized and dispatched.The present application dynamically matches the size of window and dispatch situation by analyzing the degree of urgency of each pending work order in current dispatch pool and the scale of the same kind task, adaptively adjusts the size of attention window, so as to improve the overall decision accuracy and resource utilization efficiency of property operation scheduling.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and specifically to a smart property management operation collaborative scheduling method and system based on machine learning. Background Technology

[0002] With the advancement of urbanization, the property operation of commercial complexes and large parks increasingly relies on the collaborative work of multiple professions such as engineering, cleaning, security, and customer service. Tasks in property dispatch scenarios typically have clear temporal dependencies and resource-sharing characteristics. A single facility malfunction may trigger multiple related repair work orders, and the urgency and skill requirements of different task categories also vary significantly. To address these complex scheduling needs, an attention-based dispatch scheduling method has been introduced into the field of property operation management. By weighting and prioritizing the work orders in the dispatch pool based on their features, the optimal allocation of property human resources can be achieved. The attention window, as the core computational unit of this scheduling method, directly determines the number and category range of work orders included in a single scheduling iteration, thus affecting the accuracy of task priority feature representation and the rationality of dispatch decisions.

[0003] However, existing property management work scheduling methods often employ fixed-size attention windows, which suffer from a structural mismatch between the preset window size and the actual number, category composition, and urgency of tasks in the dispatch pool. When a large number of ordinary work orders flood into the fixed window in a short period, the relative weight of the characteristics of high-priority urgent tasks within the window is diluted, resulting in urgent tasks failing to receive proper priority dispatch. When the window size is too narrow, batch-related work orders triggered by the same fault root cause are spatially fragmented into multiple scattered independent tasks, failing to trigger batch centralized handling strategies and causing maintenance personnel to make ineffective round trips between different work orders. The aforementioned failure in dynamic adaptation between the fixed window size and the actual dispatch situation causes the accuracy of task priority feature representation within the window to deteriorate significantly with changes in the dispatch situation. Summary of the Invention

[0004] This invention provides a smart property management operation collaborative scheduling method and system based on machine learning to solve the above-mentioned problems.

[0005] The present invention provides a smart property management operation collaborative scheduling method and system based on machine learning, which adopts the following technical solution:

[0006] In a first aspect, an embodiment of the present invention provides a smart property operation collaborative scheduling method based on machine learning, the method comprising: obtaining pending work orders in the current dispatch pool; evaluating the urgency of each pending work order; determining the scale of similar tasks in the pending work orders; adaptively setting the size of an attention window based on the urgency and the scale of similar tasks; prioritizing the pending work orders within the attention window and executing dispatch.

[0007] Furthermore, the adaptive setting of the attention window size based on the urgency and the scale of similar tasks includes: determining the degree of urgency deviation of each type of task and the window expansion tendency index of the overall task dispatch pool based on the changing trend of the scale of similar tasks and the urgency; determining a baseline window change factor based on the degree of urgency deviation and the window expansion tendency index; and correcting the baseline window change factor based on the trend of the continuous deterioration of urgency of each type of task to obtain the adaptively adjusted attention window size.

[0008] Furthermore, determining the degree of urgency deviation for each category of tasks and the window expansion tendency index of the overall task dispatch pool includes: aggregating and determining the window expansion tendency index of the overall task dispatch pool based on the urgency of each category of tasks; and determining the degree of urgency deviation for each category of tasks based on the changing trend of the current number of tasks relative to the number of historical batches, combined with the ratio of the urgency of the task category to the current maximum urgency.

[0009] Furthermore, the step of correcting the baseline window change factor based on the urgent and continuously deteriorating trend of each category of tasks includes: statistically analyzing the proportion of batches in which the degree of urgency deviation of each category of tasks shows an increasing trend over multiple historical periods, and the cumulative increase from the first appearance of the task of that category to the current batch; determining the urgent and continuous bias coefficient of each category of tasks based on the batch proportion and the cumulative increase; and using the maximum value among the urgent and continuous bias coefficients as a penalty term to correct the baseline window change factor.

[0010] Further, prioritizing and dispatching the pending work orders within the attention window includes: extracting task feature vectors for each pending work order within the attention window, inputting the task feature vectors into a pre-trained ranking model to obtain a ranking score for each pending work order; arranging the pending work orders within the attention window in descending order based on the ranking scores to generate a dispatch queue; and matching the dispatch queue with the real-time location, skills, and load status of property personnel to generate an optimal dispatch plan for work orders and personnel.

[0011] Furthermore, the method also includes: issuing dispatch instructions based on the optimal dispatch scheme; feeding back the execution status of each pending work order to the dispatch pool in real time; updating the training sample set based on the execution status; and continuously optimizing the sorting model based on the updated training sample set.

[0012] Furthermore, before sorting the pending work orders in the attention window in descending order based on the sorting score, the method further includes: merging similar tasks of pending work orders with the same repair category and similar facility locations in the attention window, and incorporating the merged tasks into the dispatch queue.

[0013] The step of matching the dispatch queue with the real-time location, skills, and load status of property personnel includes: generating a unified dispatch plan for similar merged tasks, and executing centralized dispatch based on the unified dispatch plan.

[0014] Furthermore, the assessment of the urgency of each pending work order includes: identifying similar tasks based on the repair category of the pending work orders; determining the degree of clustering of each category of tasks in the spatial dimension and the degree of concentration in the temporal dimension; and determining the urgency of each category of tasks based on the proportion of each category of tasks, the degree of clustering, the degree of concentration, and the scarcity of personnel to handle that category of tasks.

[0015] Furthermore, the step of obtaining pending work orders in the current dispatch pool includes: receiving original work orders from multiple sources and performing legality verification and format standardization on the original work orders; extracting repair category, facility location, repair time, and fault description keywords from the original work orders according to a predefined work order template, and mapping them to a unified data record table; associating with the historical work order database, identifying duplicate repairs and generating duplicate repair tags; based on the background skill matching results, marking the number of available personnel for the corresponding category in the unified data record table, and including the marked data record table as the pending work orders in the dispatch pool.

[0016] Secondly, another embodiment of the present invention provides a smart property management operation collaborative scheduling system based on machine learning, comprising:

[0017] The work order acquisition module is used to acquire the work orders to be processed in the current dispatch pool;

[0018] The urgency assessment module is used to assess the urgency of each of the pending work orders.

[0019] The module for determining the scale of similar tasks is used to determine the scale of similar tasks in the work order to be processed.

[0020] The window adaptive setting module is used to adaptively set the size of the attention window based on the urgency and the scale of similar tasks;

[0021] The sorting and dispatching module is used to prioritize the pending work orders within the attention window and then dispatch them.

[0022] The beneficial effects of the technical solution of the present invention are:

[0023] In this embodiment of the invention, pending work orders in the current dispatch pool are obtained; the urgency of each pending work order is evaluated; the scale of similar tasks in the pending work orders is determined; the size of the attention window is adaptively set based on the urgency and the scale of similar tasks; and the pending work orders are prioritized and dispatched within the attention window.

[0024] This invention, by analyzing the urgency of each pending work order in the current dispatch pool and the scale of similar tasks, adaptively adjusts the attention window size to dynamically match the dispatch situation, thereby improving the overall decision-making accuracy and resource utilization efficiency of property operation scheduling. On the other hand, by analyzing the spatial and temporal concentration of various task categories and the scarcity of personnel handling these tasks, the urgency of each task category is comprehensively determined, making the identification of urgent tasks no longer dependent on a single indicator, thus improving the objectivity and accuracy of priority determination. Furthermore, the invention determines urgency based on the changing trends in the scale of similar tasks and their urgency. The window changes direction and magnitude, and adjusts the window change factor when the emergency task continues to worsen. This allows the window to expand when similar tasks burst in large numbers to capture all similar task chains, and to shrink when the emergency task characteristics are prominent to focus on high-priority tasks. This avoids the dilution of emergency characteristics and the spatial fragmentation of batch failures into scattered independent tasks. On the other hand, by extracting the task feature vectors of each work order to be processed within the attention window and inputting them into a pre-trained ranking model, the optimal matching is performed by combining the real-time location, skills and load status of property personnel to generate the optimal dispatch plan for work orders and personnel, thus achieving precise docking between property human resources and task requirements. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating the machine learning-based smart property management operation collaborative scheduling method provided in this embodiment of the invention;

[0027] Figure 2 This is a schematic diagram of the architecture of a machine learning-based smart property management operation collaborative scheduling system provided in an embodiment of the present invention. Detailed Implementation

[0028] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following detailed description, in conjunction with the accompanying drawings and preferred embodiments, provides a detailed account of the specific implementation methods, structures, features, and effects of the method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0030] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent property operation collaborative scheduling method and system based on machine learning provided by this invention.

[0031] like Figure 1 As shown, this embodiment of the invention provides a smart property management operation collaborative scheduling method based on machine learning, including:

[0032] Step S110: Obtain the pending work orders in the current dispatch pool.

[0033] Optionally, step S110 above includes: receiving multiple source original work orders and performing legality verification and format standardization on the original work orders; extracting repair category, facility location, repair time, and fault description keywords from the original work orders according to a predefined work order template, and mapping them to a unified data record table; associating with the historical work order database, identifying duplicate repairs and generating duplicate repair tags; based on the background skill matching results, marking the number of available personnel for the corresponding category on the unified data record table, and including the marked data record table as a work order to be processed in the dispatch pool.

[0034] Homeowners submit repair requests via mobile apps, mini-programs, or customer service hotlines, creating initial work orders. Clients and administrators aggregate these multiple initial work orders through a unified access gateway and integrate them into the pending processing flow. The current work order pool consists of these initial work orders that have undergone the aforementioned integration process, serving as the foundational data for subsequent scheduling operations.

[0035] The data format and field integrity of original work orders from multiple sources are heterogeneous due to differences in access channels. They must undergo validity verification and format normalization before entering downstream processing stages. Validity verification screens for missing required fields, format errors, and logical inconsistencies in the original work orders, eliminating abnormal data that does not conform to access specifications. Format normalization converts the verified original work orders into a standardized format that conforms to a predefined data protocol, ensuring that work orders from different sources have a unified field structure and encoding rules.

[0036] The standardized work orders, after being formatted, enter the structured parsing engine. This engine, based on a predefined work order template, extracts the repair category, facility location, repair time, and fault description keywords from the text description and form fields. Specifically, the repair category identifies the technical job and business type of the work order; the facility location records the specific spatial coordinates or area code of the fault; the repair time marks the timestamp of the work order's creation; and the fault description keywords are extracted through semantic parsing to reveal the core characteristics of the fault.

[0037] After the data record table is generated, it is automatically linked to the historical work order database to compare the similarity between the current work order and the repair category, facility location, and fault description keywords in the historical records. When the similarity exceeds a preset threshold, the current work order is identified as a duplicate repair and a duplicate repair tag is generated. This tag can reflect the situation where the same facility or similar faults are reported multiple times in a short period of time, providing a historical reference for subsequent urgency assessment and aggregation of similar tasks.

[0038] The structured work order record table enters the work order pool queue. The subsequent dispatch decision-making process matches the record table against the background skills database, marking the number of available personnel with the skills to handle the corresponding repair category. The number of available personnel reflects the current human resource scale capable of handling this type of task and is a key parameter for measuring task processing bottlenecks and personnel scarcity, directly influencing subsequent urgency scoring and adaptive window size adjustment.

[0039] Step S120: Assess the urgency of each pending work order.

[0040] It is understandable that in property management collaborative scheduling scenarios, pending work orders exhibit significant heterogeneity in terms of repair category, fault severity, and response time requirements. Without a quantitative representation of the urgency of each task category, the task priority characteristics within a fixed attention window will be diluted by fluctuations in the number of ordinary work orders, resulting in high-priority urgent tasks failing to obtain a relative ranking weight commensurate with their urgency attributes within the window. By assessing the urgency of each pending work order, the task category with the highest response time requirement in the current dispatch pool can be identified, providing directional guidance for the adaptive adjustment of the attention window size. When high-urgency task categories dominate, the window should shrink to focus on high-priority tasks, preventing their characteristics from being overwhelmed by a large number of ordinary work orders. Simultaneously, the urgency assessment results also form the quantitative basis for subsequent work order priority ranking and dispatch decisions within the window, enabling the ranking model to differentiate tasks based on objectively calculated urgency indices, rather than relying on static rules or fixed thresholds. Therefore, urgency assessment is a crucial intermediate link connecting dispatch pool status awareness and window adaptive adjustment, directly determining the accuracy of the scheduling decision's response to real-time changes in the dispatch situation. The following describes one method for determining the urgency of pending work orders:

[0041] Optionally, step S120 above includes: identifying similar tasks for work orders to be processed according to repair categories; determining the degree of clustering of each category of tasks in the spatial dimension and the degree of concentration in the temporal dimension; and determining the urgency of each category of tasks based on the proportion of each category of tasks, the degree of clustering, the degree of concentration, and the scarcity of personnel to handle that category of tasks.

[0042] The pending work orders in the property management dispatch pool naturally exhibit clustering characteristics based on repair category. The same facility malfunction or event in the same area often triggers multiple repair requests with similar technical attributes. Aggregating and identifying work orders by repair category divides the work order set into... Mutually exclusive subsets of categories Each category Include Individual work orders. This classification method is not a simple labeling, but rather establishes statistical units based on categories for subsequent batch centralized analysis in spatial and temporal dimensions. This allows independently entered work orders to be reconstructed into a batch task chain triggered by the same fault root cause, providing a data organization basis for determining whether there is physical feasibility for centralized handling.

[0043] Building upon the identification of similar tasks, it is necessary to further quantify the spatial distribution density of each task category within the facility. For each category... Any two work orders within and Calculate the Euclidean distance between the locations of its facilities. The arithmetic mean of the distances between all pairwise combinations within the category is taken to obtain the category average distance. Then calculate all orders in the current dispatch pool. Mean distance of each category , serving as a global benchmark for spatial distribution. The spatial aggregation factor of a class is calculated using the following formula:

[0044]

[0045] In the formula, To prevent division by zero of extremely small positive numbers; when much smaller hour, Approaching 1 indicates that this type of task is highly concentrated in space; when Much larger hour, A value approaching 0 indicates a high degree of spatial dispersion. The aforementioned spatial clustering factor transforms absolute spatial distance into a normalized clustering measure relative to the global mean, eliminating the dimensional effects caused by differences in the scale of different property areas.

[0046] Besides spatial distribution, the overlap in reporting times for similar tasks directly reflects the synchronicity of fault outbreaks. For categories... Any two work orders within and Calculate the absolute difference between the timestamps of the repair requests. The arithmetic mean of the time differences for all pairwise combinations within the category is taken to obtain the category average time difference. Then calculate the mean time difference for all categories. , serving as a global benchmark for time distribution. The time concentration factor of a class is calculated using the following formula:

[0047]

[0048] in, To prevent division by zero of extremely small positive numbers; when much smaller hour, A value approaching 1 indicates a high concentration of repair requests; when Much larger hour, A value approaching 0 indicates a high degree of temporal dispersion. The temporal concentration factor and the spatial concentration factor together constitute the dual criteria for batch concentration; neither can be dispensed with. Tasks that are only spatially similar but have a large time span are usually independent faults and do not meet the conditions for centralized handling.

[0049] Spatial clustering factor and temporal clustering factor characterize the physical feasibility of batch centralized processing, but they do not yet reflect the relative size weight of such tasks in the overall dispatch pool. Introducing a batch clustering index:

[0050]

[0051] In the formula, For category The number of sub-work orders, This represents the total number of work orders in the current work order pool. and These are the spatial clustering factor and the temporal clustering factor corresponding to this category, respectively. It can measure the proportion of similar tasks among all pending work orders. The product of these three factors—scale effect and clustering effect—measures the concentration of this type of task across time and space. For all... Calculation by category Then, take their arithmetic mean. This serves as a benchmark for measuring the overall batch concentration of the current order dispatch pool.

[0052] The batch centralization index characterizes the physical feasibility of centrally handling similar tasks, but fails to reflect the impact of human resource bottlenecks on the urgency of deployment. The total number of currently deployable personnel is denoted as... It has processing categories The number of people who can use the skill is denoted as .ratio This characterizes the relative scarcity of the skills required to perform this type of task. The smaller the value, the more severe the human resource bottleneck. Combining the relative deviation between the skill scarcity factor and the batch concentration index yields the initial urgency score:

[0053]

[0054] In the formula, Characterizes the deviation of the current category's batch concentration from the global average level; when If no matching personnel are available, the urgency level for that category should be set to 0 to avoid inclusion in invalid scheduling. (Original score) For dimensionless parameters, the normalization method is used to map them to... The interval is used to obtain the final urgency index. . The closer it is to 1, the more serious the backlog of this type of task is in the current task pool and the more scarce the manpower is. The more necessary it is to expand the attention window to capture more chains of similar tasks at once. The closer it is to -1, the more dispersed the tasks are and the more manpower is available, resulting in a lower demand for window expansion.

[0055] Step S130: Determine the scale of similar tasks in the pending work orders.

[0056] The implementation of step S130 above includes, for example, aggregating and identifying pending work orders in the current dispatch pool according to repair categories, grouping work orders with the same repair category into the same category subset, and counting the number of sub-work orders in each category subset and their proportion to the total number of work orders in the current dispatch pool to obtain the relative scale weight of each category of tasks. Based on this, work orders with the same repair category, similar facility locations, and approximately overlapping repair times are merged into a single task chain, restoring multiple independent repair orders caused by the same fault root cause to a single task chain, eliminating the artificially inflated scale caused by repeated entry of the same event, thereby determining the actual scale of similar tasks. The proportional relationship between this actual scale and the current number of dispatchable personnel directly reflects the degree of manpower shortage for similar tasks, providing a quantitative basis for the adaptive adjustment of the attention window size.

[0057] Step S140: Adaptively set the attention window size based on urgency and the scale of similar tasks.

[0058] Optionally, step S140 includes: determining the degree of urgency deviation of each type of task and the window expansion tendency index of the overall task dispatch pool based on the changing trend of the scale of similar tasks and their urgency; determining the baseline window change factor based on the degree of urgency deviation and the window expansion tendency index; and correcting the baseline window change factor based on the trend of continuous deterioration of the urgency of each type of task to obtain the adaptively adjusted attention window size.

[0059] Property operation scheduling typically uses a preset base window size as the calculation benchmark for a single scheduling iteration. This represents the upper limit of the number of work orders that can be included in the field of view at a single time under normal dispatch conditions. However, the fixed basic window cannot adapt to the real-time fluctuations in the number, category composition, and urgency of tasks within the dispatch pool, so a dynamic adjustment mechanism needs to be introduced. By aggregating the urgency index of all categories of tasks in the current dispatch pool, the overall dispatch pool's window expansion tendency index is obtained to quantify the overall demand direction for window size adjustment under the current dispatch situation.

[0060] Optionally, the determination of the urgency deviation of each type of task and the expansion tendency index of the overall task dispatch pool includes: determining the expansion tendency index of the overall task dispatch pool based on the urgency of each type of task; and determining the urgency deviation of each type of task based on the changing trend of the current number of tasks relative to the number of historical batches, combined with the ratio of the urgency of the task category to the current maximum urgency.

[0061] Overall order pool expansion tendency index Calculate using the following formula:

[0062]

[0063] In the formula, This represents the total number of categories in the current dispatch pool. For the first The urgency index of the task type is normalized and mapped to Interval. The value is centered around 1, when the urgency of tasks in all categories is generally high. A value approaching 2 indicates a strong need for window expansion in the current task dispatch pool, requiring a larger window to capture more similar task chains; when urgency is generally low... A value approaching 0 indicates a high necessity for window contraction. Overall single-pool window expansion tendency index. It aggregates scattered category-level urgency information into a global window adjustment signal, but does not yet consider dynamic features such as sudden changes in quantity and continuous deterioration.

[0064] To characterize the degree of deviation of each task category from its historical state at the current moment, an emergency deviation coefficient is introduced. Let the current batch category be... The number of subtasks is The number of subtasks of this type that appeared in the window at the previous moment was The maximum urgency level for each task category in the current task dispatch pool is: , No. The urgency deviation coefficient for a task type is calculated using the following formula:

[0065]

[0066] In the formula, To prevent extremely small positive numbers from being divided by zero; It can characterize the relative position of the current category's urgency within the global urgency spectrum; It can characterize the quantity trend of the current batch relative to the previous batch; when a certain type of task has both high urgency and a sudden increase in quantity, If the number of tasks increases significantly, it indicates an urgent surge in orders, and the window for expansion can be widened accordingly; if the number of tasks increases but the urgency level is low, then... The deviation is lowered, thus limiting the window widening range. (The aforementioned emergency deviation coefficient...) This allows the window to adjust not only in response to static urgency but also to capture dynamic changes in the situation caused by sudden shifts in quantity.

[0067] Optionally, the above-mentioned correction of the baseline window change factor based on the urgent and continuously deteriorating trend of each type of task includes: statistically analyzing the proportion of batches in which the degree of urgent deviation of each type of task shows an increasing trend over multiple historical periods and the cumulative increase from the first appearance of the task of that type to the current batch; determining the urgent and continuous bias coefficient of each type of task based on the batch proportion and the cumulative increase; and using the maximum value among the urgent and continuous bias coefficients as a penalty term to correct the baseline window change factor.

[0068] In real-world property management scenarios, certain tasks may maintain a high level of urgency and exhibit a worsening trend across multiple batches, such as repeated pipe bursts due to aging pipes. In such cases, it's necessary to track the changes in the urgency deviation coefficient over multiple historical periods to determine the continuously deteriorating nature of the urgency level. Let's define categories. A total of records have been made from the first appearance to the current moment. Each batch The value, of which the set of times that satisfy the growth compared to the previous batch contains a total of For each growth moment, define the emergency persistence bias coefficient:

[0069]

[0070] In the formula, The total number of comparable adjacent batch pairs; For the current batch (time) The emergency deviation coefficient, This is the emergency deviation coefficient when this type of task first appears. It can characterize the proportion of moments of growth and reflect the frequency dimension of a continuous increase in the level of urgency; It can characterize the cumulative growth rate from the first appearance to the current batch, reflecting the severity of the deterioration. The product of the two factors makes... Simultaneously considering both sustainability and the extent of deterioration, when a certain type of task has historically increased multiple times and generally shows an upward trend... A significant increase indicates a strong and persistent bias.

[0071] When determining the direction of change of the baseline window, the window expansion tendency index should be used. Based on this, and corrected using the maximum value in the emergency persistence bias coefficient as a penalty term. Window variation factor Calculate using the following formula:

[0072]

[0073] In the formula, This is the maximum value among all categories of urgency persistence bias coefficients. When the urgency of a certain type of task continues to deteriorate (… ), The window is compressed, narrowing its expansion range, allowing it to actively focus on the continuously worsening task and preventing a large influx of routine work orders from diluting its urgency. Applying the change factor to the base window yields the adaptively adjusted actual window size:

[0074]

[0075] In the formula, The attention window size is adaptively adjusted; This is the preset base window size. Follow The window adjusts in real time according to the dynamic changes, so that when a batch of similar tasks bursts out, the window expands to capture all similar task chains, and shrinks to maintain its relative ranking weight when urgent tasks continue to worsen, thus achieving a precise balance between the needs of expanding and shrinking the window.

[0076] The aforementioned adaptive window adjustment mechanism captures the coupling characteristics of sudden changes in quantity and urgency at the current moment through the emergency deviation coefficient, and tracks historical deterioration trends through the emergency persistence bias coefficient. These two mechanisms complement the window expansion tendency index. The window expansion tendency index provides a global adjustment benchmark, the emergency deviation coefficient characterizes instantaneous deviation, and the emergency persistence bias coefficient imposes a continuous penalty. This three-layer progressive structure ensures that the window size is no longer limited to a fixed preset value, but dynamically adapts to the real-time evolution of the dispatch situation. This mechanism ensures that when similar tasks accumulate rapidly in a short period due to the same root cause of the fault, the window expands to include all similar work orders in the same view to trigger batch centralized processing. When the urgency of a certain type of task continues to rise in multiple batches, the window change factor automatically shrinks, maintaining a high ranking weight for that type of task within the window and prioritizing access to scarce maintenance resources.

[0077] Step S150: Prioritize the work orders to be processed within the attention window and execute the dispatch.

[0078] Optionally, step S150 includes: extracting the task feature vector of each work order to be processed within the attention window, and inputting the task feature vector into a pre-trained ranking model to obtain the ranking score of each work order to be processed; arranging the work orders to be processed within the attention window in descending order based on the ranking score to generate a dispatch queue; and matching the dispatch queue with the real-time location, skills, and load status of property personnel to generate the optimal dispatch plan for work orders and personnel.

[0079] After adaptive adjustment, the attention window captures a set of work orders that needs to be transformed into a numerical representation that can be used by the ranking model. The task feature vectors extracted from each work order to be processed cover dimensions such as repair category, facility location, urgency index, and batch concentration index, which correspond to the technical job attributes, spatial coordinate information, time urgency, and feasibility assessment of centralized handling of similar tasks, respectively.

[0080] After the aforementioned feature vectors are input into a pre-trained ranking model, the model outputs a relative ranking score for each work order to be processed. This score is not a scalar measure of absolute priority, but rather an ordinal value representing the relative priority relationship between tasks within the current window. In an optional implementation, the ranking model can employ a gradient boosting tree-based ranking learner. By integrating multiple decision trees to non-linearly partition the feature space, it captures the interaction effect between urgency and batch concentration index, thereby outputting a score estimate reflecting the overall scheduling priority. A higher score indicates a greater dispatch priority that the task should receive in the current scheduling iteration.

[0081] Based on the scores output by the ranking model, all pending work orders within the attention window are sorted in descending order to generate a priority dispatch queue. This queue linearly organizes the scattered work orders within the window according to their overall priority from high to low, ensuring that high-urgency tasks and similar tasks with batch centralized processing conditions can enter the subsequent personnel matching stage first. The result of the descending order directly determines the processing sequence of each work order in the current scheduling cycle, serving as a crucial intermediate link connecting task priority assessment and human resource allocation. The ranking model can employ XGBoost Ranker (a ranking model based on XGBoost). This model takes the work order set after adaptive window truncation as input, extracts feature vectors of work order category, facility location, urgency index, and batch centralized index for each task unit, and outputs the relative ranking score of each task unit within the current window. A higher score indicates a higher overall dispatch priority for the task. Based on this, all tasks within the window are sorted in descending order to form the final priority dispatch queue. It should be noted that the aforementioned XGBoost Ranker model is a mature model in existing technology. For its architecture and working principle, please refer to relevant technologies; the embodiments of this invention will not elaborate further.

[0082] After the dispatch queue is generated, it needs to be matched with the real-time status of property personnel to generate an executable dispatch plan. The real-time status of property personnel includes their current geographical location, processing skills, and current task load. These three factors jointly constrain the feasibility and efficiency of personnel accepting new work orders. In one optional implementation, an optimal bipartite graph matching algorithm can be used to globally optimize the task demand and personnel supply. Under the premise of satisfying skill matching constraints, the algorithm optimizes the objective functions such as spatial distance and load balancing between work orders and personnel, thereby outputting the optimal dispatch plan for work orders and personnel. The Hungarian algorithm is a typical optimal bipartite graph matching algorithm. In this embodiment of the invention, the dispatch queue, after being sorted in descending priority, constitutes the task-side vertex set of the bipartite graph, and the real-time location, skills, and current load information of property personnel constitute the personnel-side vertex set. The edge weights between the two vertices comprehensively reflect spatial accessibility, skill matching degree, and load balancing. The Hungarian algorithm iteratively finds augmenting paths and solves the maximum weight matching that maximizes the sum of the overall edge weights under the skill matching constraints, outputting the optimal dispatch plan between work orders and property personnel, thus achieving a global optimization of the allocation of task demand and human resource supply. It should be noted that the above-mentioned Hungarian algorithm is a mature and optimal bipartite graph matching algorithm in the prior art. For its implementation and working principle, please refer to the prior art. The embodiments of this invention will not be repeated here.

[0083] Optionally, before sorting the pending work orders in the attention window in descending order based on the sorting score, the method further includes: merging the pending work orders with the same repair category and similar facility locations in the attention window, and including the merged tasks in the dispatch queue.

[0084] The above-mentioned matching of dispatch queues with the real-time location, skills, and load status of property personnel includes: generating a unified dispatch plan for similar tasks after merging, and executing centralized dispatch based on the unified dispatch plan.

[0085] After adaptive adjustment, the attention window may contain multiple work orders with the same repair category and similar facility locations. These work orders essentially originate from the same root cause of the fault, but were entered separately by the owner as independent work orders during the repair process. Directly sorting and dispatching these work orders individually would result in inefficient back-and-forth trips between nearby facility locations by the same maintenance personnel, failing to leverage the efficiency advantages of centralized batch processing. Therefore, before sorting in descending order based on ranking scores, a preprocessing step of merging similar work orders within the attention window can be performed. This merges multiple work orders with the same repair category and facility locations meeting a preset proximity threshold into a single merged task unit, restoring the scattered independent repair orders into a task chain capable of centralized processing.

[0086] The criteria for merging similar tasks encompass both the consistency of the repair category and the proximity of the facility location. The same repair category is a fundamental prerequisite for merging, ensuring that the merged task unit corresponds to a single technical job and processing skill. Facility proximity is quantified using a spatial distance threshold. When the Euclidean distance between the facility locations of multiple similar work orders is less than a preset threshold, they are deemed to meet the spatial proximity condition and can be included in the same merged task unit. The merged task unit retains the key attribute information of each original sub-work order. For example, the urgency level of the merged unit is taken as the maximum or weighted average of the urgency levels of each sub-work order, and the facility location is taken as the centroid coordinates of the locations of each sub-work order. This generates a single comprehensive task object representing the batch of tasks, replacing the original multiple scattered independent work orders in the subsequent sorting process.

[0087] Within the attention window after merging similar tasks, the set of work orders to be processed consists of both independent work orders and merged task units. Task feature vectors are extracted from the set of work orders to be processed and input into the ranking model, which outputs a ranking score for each task unit. The ranking model maintains consistency with the processing of independent work orders in terms of feature extraction and score calculation for merged task units. Features such as the overall urgency and batch concentration index of merged task units have already been aggregated in the preceding merging steps; therefore, the ranking model can directly output a global priority sequence containing merged task units without additional adaptation. After being sorted in descending order based on the ranking scores, merged task units occupy a position in the dispatch queue corresponding to their overall priority, participating in the subsequent personnel matching process along with independent work orders.

[0088] When matching the dispatch queue with the real-time status of property personnel, the merged task unit participates in the bipartite graph matching operation as a single vertex. During the generation of the optimal dispatch plan, the matching algorithm treats the merged task unit as an indivisible scheduling object, assigning it a single property personnel member or a team of personnel with the corresponding skills and within the current load limits. A unified dispatch plan is generated for merged tasks of the same type. This plan includes the processing order of all sub-work orders within the merged task unit, the estimated arrival time, and a list of required tools and spare parts, enabling maintenance personnel to complete the processing of all sub-work orders within the batch task chain in one go, avoiding multiple round trips. Centralized dispatch is executed based on the unified dispatch plan. Maintenance personnel process each sub-work order within the merged task unit sequentially according to the planned path and order. After processing, the execution status is uniformly fed back to the dispatch pool, forming a closed-loop iteration.

[0089] Optionally, the above-mentioned machine learning-based smart property operation collaborative scheduling method further includes: issuing dispatch instructions based on the optimal dispatch plan; feeding back the execution status of each pending work order to the dispatch pool in real time; updating the training sample set based on the execution status; and continuously optimizing the sorting model based on the updated training sample set.

[0090] After the optimal dispatch plan is generated, it can be parsed into a dispatch instruction and sent to the mobile terminal of the corresponding property staff via the dispatch interface. The instruction includes work order details, facility location navigation information, and estimated processing time. After receiving the instruction, the property staff will go to the site to execute it. The status of key nodes in the execution process (including order confirmation, arrival check-in, processing, and processing completion) will be transmitted back to the dispatch pool in real time, thereby continuously monitoring the lifecycle status of each work order.

[0091] After the execution status is returned to the dispatch pool, in addition to supporting status monitoring for the current scheduling cycle, the execution status can also serve as real feedback data for ranking model optimization. Incremental training samples are generated based on the deviation between the actual execution results and the model's predicted ranking. Each sample record contains the feature vector of each work order within the window, the ranking score output by the model, and the real priority label verified after actual execution (such as actual processing time, owner satisfaction, or fault impact range), forming a complete training record from feature input to predicted output to real feedback.

[0092] Based on a continuously accumulating incremental training sample set, the ranking model periodically performs incremental updates or online learning. During the update process, the model aims to minimize the loss function between the predicted ranking and the actual feedback, adjusting the feature weights and decision boundaries to make the ranking scores output in subsequent scheduling iterations more closely reflect the efficiency and service quality requirements of actual property operations.

[0093] like Figure 2 As shown, this embodiment of the invention also provides a smart property management operation collaborative scheduling system 200 based on machine learning, comprising:

[0094] Work order acquisition module 210 is used to acquire pending work orders in the current dispatch pool;

[0095] The urgency assessment module 220 is used to assess the urgency of each pending work order;

[0096] The module 230 for determining the scale of similar tasks is used to determine the scale of similar tasks in the work order to be processed.

[0097] The window adaptive setting module 240 is used to adaptively set the size of the attention window based on urgency and the scale of similar tasks.

[0098] The sorting and dispatching module 250 is used to prioritize and dispatch work orders within the attention window.

[0099] It should be noted that the above-mentioned machine learning-based smart property operation collaborative scheduling system 200 can realize any one of the functions of the machine learning-based smart property operation collaborative scheduling method provided in the embodiments of the present invention. For the method embodiment section, please refer to the method embodiment section for the way each function is realized and the working principle. The system embodiment section will not repeat it here.

[0100] This invention is now complete.

[0101] In summary, in this embodiment of the invention, the following steps are taken: 1) Obtain pending work orders in the current dispatch pool; 2) Assess the urgency of each pending work order; 3) Determine the scale of similar tasks among the pending work orders; 4) Adaptively set the attention window size based on the urgency and scale of similar tasks; 5) Prioritize the pending work orders within the attention window and execute dispatch.

[0102] This invention analyzes the urgency of each pending work order in the current dispatch pool and the scale of similar tasks, adaptively adjusting the attention window size to dynamically match the dispatch situation, thereby improving the overall decision-making accuracy and resource utilization efficiency of property operation scheduling. On the other hand, by analyzing the spatial and temporal clustering of various task categories and the scarcity of personnel handling these tasks, the urgency of each task category is comprehensively determined, making the identification of urgent tasks no longer reliant on a single indicator, thus improving the objectivity and accuracy of priority determination. Furthermore, the attention window size is determined based on the changing trends in the scale of similar tasks and their urgency. The system adjusts the direction and magnitude of window changes and modifies the window change factor when emergency tasks continue to worsen. This allows the window to expand when similar tasks burst in large numbers to capture all similar task chains, and to contract when emergency task characteristics are prominent to focus on high-priority tasks. This avoids diluting emergency characteristics and spatially segmenting batch failures into scattered independent tasks. On the other hand, by extracting the task feature vectors of each work order to be processed within the attention window and inputting them into a pre-trained ranking model, the system performs optimal matching based on the real-time location, skills, and load status of property personnel to generate the optimal dispatch plan for work orders and personnel, thus achieving precise matching between property human resources and task requirements.

[0103] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart property management operation collaborative scheduling method based on machine learning, characterized in that, The method includes: Retrieve pending work orders from the current dispatch pool; Assess the urgency of each pending work order; Determine the scale of similar tasks in the pending work orders; Based on the urgency and the scale of similar tasks, the attention window size is adaptively set; Prioritize the pending work orders within the attention window and then dispatch them.

2. The intelligent property management operation collaborative scheduling method based on machine learning according to claim 1, characterized in that, The adaptive setting of the attention window size based on the urgency and the scale of similar tasks includes: Based on the changing trend of the scale of similar tasks and the urgency, the degree of urgency deviation of each type of task and the window expansion tendency index of the overall task dispatch pool are determined. Based on the degree of emergency deviation and the window expansion tendency index, a baseline window change factor is determined; The baseline window change factor is modified based on the urgent and continuous deterioration trend of each type of task to obtain the adaptively adjusted attention window size.

3. The intelligent property management operation collaborative scheduling method based on machine learning according to claim 2, characterized in that, The determination of the urgency deviation degree of each type of task and the windowing tendency index of the overall dispatch pool includes: Based on the urgency of each type of task, the overall order dispatch pool's window expansion tendency index is determined by aggregation. Based on the changing trend of the current number of tasks in each category relative to the number of historical batches, and combined with the ratio of the urgency of the task category to the current maximum urgency, the degree of urgency deviation of each category of tasks is determined.

4. The intelligent property management operation collaborative scheduling method based on machine learning according to claim 2, characterized in that, The correction of the baseline window change factor based on the urgent and continuous deterioration trend of each type of task includes: The percentage of batches where the degree of urgency deviation for each type of task showed an increasing trend over multiple historical periods was statistically analyzed, along with the cumulative increase from the first appearance of that type of task to the current batch. Based on the batch proportion and the cumulative growth rate, the emergency sustainability bias coefficient for each type of task is determined; The maximum value among the aforementioned emergency persistence bias coefficients is used as a penalty term to correct the baseline window change factor.

5. The intelligent property management operation collaborative scheduling method based on machine learning according to claim 1, characterized in that, The step of prioritizing and dispatching the pending work orders within the attention window includes: Extract the task feature vector of each work order to be processed within the attention window, and input the task feature vector into the pre-trained ranking model to obtain the ranking score of each work order to be processed. Based on the sorting score, the pending work orders in the attention window are sorted in descending order to generate a dispatch queue; The dispatch queue is matched with the real-time location, skills, and load status of property personnel to generate the optimal dispatch plan for work orders and personnel.

6. The intelligent property management operation collaborative scheduling method based on machine learning according to claim 5, characterized in that, The method further includes: Based on the optimal dispatch scheme, issue a dispatch instruction; The execution status of each pending work order is fed back to the dispatch pool in real time; The training sample set is updated based on the execution state, and the ranking model is continuously optimized based on the updated training sample set.

7. The intelligent property management operation collaborative scheduling method based on machine learning according to claim 5, characterized in that, Before sorting the pending work orders within the attention window in descending order based on the sorting score, the method further includes: For pending work orders with the same repair category and similar facility location within the attention window, merge similar tasks and include the merged tasks in the dispatch queue. The step of matching the dispatch queue with the real-time location, skills, and workload of property management personnel includes: A unified dispatch plan is generated for the merged tasks of the same type, and centralized dispatch is executed based on the unified dispatch plan.

8. The intelligent property management operation collaborative scheduling method based on machine learning according to any one of claims 1-7, characterized in that, The assessment of the urgency of each pending work order includes: The pending work orders are identified as similar tasks according to the repair category; Determine the degree of clustering of each type of task in the spatial dimension and the degree of concentration in the temporal dimension; The urgency of each task category is determined based on the proportion of each category, the degree of clustering, the degree of concentration, and the scarcity of personnel to handle that category of task.

9. The intelligent property management operation collaborative scheduling method based on machine learning according to any one of claims 1-7, characterized in that, The step of obtaining the pending work orders in the current dispatch pool includes: Receive original work orders from multiple sources and perform legality verification and format standardization on the original work orders; Based on the predefined work order template, the repair category, facility location, repair time, and fault description keywords are extracted from the original work order and mapped to a unified data record table; Link to the historical work order database to identify duplicate repair requests and generate duplicate repair flags; Based on the background skill matching results, the number of available personnel for the corresponding category is marked on the unified data record table, and the marked data record table is included in the dispatch pool as the work order to be processed.

10. A smart property management operation collaborative scheduling system based on machine learning, characterized in that, include: The work order acquisition module is used to acquire the work orders to be processed in the current dispatch pool; The urgency assessment module is used to assess the urgency of each of the pending work orders. The module for determining the scale of similar tasks is used to determine the scale of similar tasks in the work order to be processed. The window adaptive setting module is used to adaptively set the size of the attention window based on the urgency and the scale of similar tasks; The sorting and dispatching module is used to prioritize the pending work orders within the attention window and then dispatch them.

Images